Haiyu Wu

Haiyu Wu

Research Scientist, Altos Labs

hwu[at]altoslabs[dot]com

About Me

I am a Research Scientist at Altos Labs. Currently, I am working on building world models for capturing the essential pattern of biological data Before this, I earned my PhD degree at the University of Notre Dame, where I worked with Prof. Kevin W. Bowyer at the Computer Vision Research Lab (CVRL).

The long-term goal of my research is to improve the transfer learning, zero-shot, and few-shot capabilities of foundation models, and to understand how models can effectively learn useful knowledge from noisy data.

Research Interests

  • World Model
  • Multi-modal
  • Effective Learning from Noisy Data

Education

  • PhD in Computer Science and Engineering
    University of Notre Dame

News

Experience

Present
Research Scientist Altos Labs

Leading a team to build a world model for capturing essence of biology from low-quality data.

2022 – 2025
Graduate Research Assistant University of Notre Dame

Achieving identity-controlled large-scale face generation dataset generation (thesis link). I am the first one showing images can be directly generated from representation features (Vec2Face and Vec2Face+).

Featured Publications

Goldilocks Test Sets for Face Verification

Haiyu Wu, Sicong Tian, Aman Bhatta, Jacob Gutierrez, Grace Bezold, Genesis Argueta, Michael C. King, Kevin W. Bowyer

CVPR

We introduce a new protocol for assembling face verification test sets and provide three valuable commercial level test sets to the community.

Vec2Face+ for Face Dataset Generation

Haiyu Wu, Jaskirat Singh, Sicong Tian, Liang Zheng, Kevin W. Bowyer

arXiv

Attribute-controlled version of Vec2Face. First synthetic face dataset that achieves a higher accuracy than real-world datasets.

Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors

Haiyu Wu, Jaskirat Singh, Sicong Tian, Liang Zheng, Kevin W. Bowyer

ICLR

Generating face images directly from face representation vectors. Easy to scale and fast to generate.

LogicNet: A Logical Consistency Embedded Face Attribute Learning Network

Haiyu Wu, Haiyu Wu, Sicong Tian, Huayu Li, Kevin W. Bowyer

WACV

Propose a adversarial training framework to resolve the logical inconsistency problem in multi-attribute classification.

Logical Consistency and Greater Descriptive Power for Facial Hair Attribute Learning

Haiyu Wu, Grace Bezold, Aman Bhatta, Kevin W. Bowyer

CVPR

Revealing that existing multi-attribute classification methods cannot preserve logical consistency.

What Should Be Balanced in a “Balanced” Face Recognition Dataset?

Haiyu Wu, Kevin W. Bowyer

BMVC

Investigating what demographic attributes should be balanced in a "balanced" face recognition dataset.